Scene Understanding and Semantic Retrieval for Industrial Safety

This is an industrial AI video analysis system built for TechnipFMC. The project develops a vision-language pipeline for object detection and automated captioning on industrial video footage, generating natural language descriptions of video segments. To allow querying for videos using free-form keywords, an embedding-based semantic search was implemented to a web interface. It includes embedding-based semantic search so safety analysts can query video content using free-form text rather than exact keywords. A full-stack web application exposes the retrieval system, and a benchmarking suite evaluates model safety awareness against the iSafetyBench dataset.

Видео Scene Understanding and Semantic Retrieval for Industrial Safety канала D2K Lab Rice University